1 /* SPDX-License-Identifier: GPL-2.0 */
3 * Copyright (c) 2011-2014, Intel Corporation.
9 #include <linux/nvme.h>
10 #include <linux/cdev.h>
11 #include <linux/pci.h>
12 #include <linux/kref.h>
13 #include <linux/blk-mq.h>
14 #include <linux/sed-opal.h>
15 #include <linux/fault-inject.h>
16 #include <linux/rcupdate.h>
17 #include <linux/wait.h>
18 #include <linux/t10-pi.h>
19 #include <linux/ratelimit_types.h>
21 #include <trace/events/block.h>
23 extern const struct pr_ops nvme_pr_ops;
25 extern unsigned int nvme_io_timeout;
26 #define NVME_IO_TIMEOUT (nvme_io_timeout * HZ)
28 extern unsigned int admin_timeout;
29 #define NVME_ADMIN_TIMEOUT (admin_timeout * HZ)
31 #define NVME_DEFAULT_KATO 5
33 #ifdef CONFIG_ARCH_NO_SG_CHAIN
34 #define NVME_INLINE_SG_CNT 0
35 #define NVME_INLINE_METADATA_SG_CNT 0
37 #define NVME_INLINE_SG_CNT 2
38 #define NVME_INLINE_METADATA_SG_CNT 1
42 * Default to a 4K page size, with the intention to update this
43 * path in the future to accommodate architectures with differing
44 * kernel and IO page sizes.
46 #define NVME_CTRL_PAGE_SHIFT 12
47 #define NVME_CTRL_PAGE_SIZE (1 << NVME_CTRL_PAGE_SHIFT)
49 extern struct workqueue_struct *nvme_wq;
50 extern struct workqueue_struct *nvme_reset_wq;
51 extern struct workqueue_struct *nvme_delete_wq;
52 extern struct mutex nvme_subsystems_lock;
55 * List of workarounds for devices that required behavior not specified in
60 * Prefers I/O aligned to a stripe size specified in a vendor
61 * specific Identify field.
63 NVME_QUIRK_STRIPE_SIZE = (1 << 0),
66 * The controller doesn't handle Identify value others than 0 or 1
69 NVME_QUIRK_IDENTIFY_CNS = (1 << 1),
72 * The controller deterministically returns 0's on reads to
73 * logical blocks that deallocate was called on.
75 NVME_QUIRK_DEALLOCATE_ZEROES = (1 << 2),
78 * The controller needs a delay before starts checking the device
79 * readiness, which is done by reading the NVME_CSTS_RDY bit.
81 NVME_QUIRK_DELAY_BEFORE_CHK_RDY = (1 << 3),
84 * APST should not be used.
86 NVME_QUIRK_NO_APST = (1 << 4),
89 * The deepest sleep state should not be used.
91 NVME_QUIRK_NO_DEEPEST_PS = (1 << 5),
94 * Problems seen with concurrent commands
96 NVME_QUIRK_QDEPTH_ONE = (1 << 6),
99 * Set MEDIUM priority on SQ creation
101 NVME_QUIRK_MEDIUM_PRIO_SQ = (1 << 7),
104 * Ignore device provided subnqn.
106 NVME_QUIRK_IGNORE_DEV_SUBNQN = (1 << 8),
109 * Broken Write Zeroes.
111 NVME_QUIRK_DISABLE_WRITE_ZEROES = (1 << 9),
114 * Force simple suspend/resume path.
116 NVME_QUIRK_SIMPLE_SUSPEND = (1 << 10),
119 * Use only one interrupt vector for all queues
121 NVME_QUIRK_SINGLE_VECTOR = (1 << 11),
124 * Use non-standard 128 bytes SQEs.
126 NVME_QUIRK_128_BYTES_SQES = (1 << 12),
129 * Prevent tag overlap between queues
131 NVME_QUIRK_SHARED_TAGS = (1 << 13),
134 * Don't change the value of the temperature threshold feature
136 NVME_QUIRK_NO_TEMP_THRESH_CHANGE = (1 << 14),
139 * The controller doesn't handle the Identify Namespace
140 * Identification Descriptor list subcommand despite claiming
141 * NVMe 1.3 compliance.
143 NVME_QUIRK_NO_NS_DESC_LIST = (1 << 15),
146 * The controller does not properly handle DMA addresses over
149 NVME_QUIRK_DMA_ADDRESS_BITS_48 = (1 << 16),
152 * The controller requires the command_id value be limited, so skip
153 * encoding the generation sequence number.
155 NVME_QUIRK_SKIP_CID_GEN = (1 << 17),
158 * Reports garbage in the namespace identifiers (eui64, nguid, uuid).
160 NVME_QUIRK_BOGUS_NID = (1 << 18),
163 * No temperature thresholds for channels other than 0 (Composite).
165 NVME_QUIRK_NO_SECONDARY_TEMP_THRESH = (1 << 19),
168 * Disables simple suspend/resume path.
170 NVME_QUIRK_FORCE_NO_SIMPLE_SUSPEND = (1 << 20),
173 * MSI (but not MSI-X) interrupts are broken and never fire.
175 NVME_QUIRK_BROKEN_MSI = (1 << 21),
178 * Align dma pool segment size to 512 bytes
180 NVME_QUIRK_DMAPOOL_ALIGN_512 = (1 << 22),
184 * Common request structure for NVMe passthrough. All drivers must have
185 * this structure as the first member of their request-private data.
187 struct nvme_request {
188 struct nvme_command *cmd;
189 union nvme_result result;
194 #ifdef CONFIG_NVME_MULTIPATH
195 unsigned long start_time;
197 struct nvme_ctrl *ctrl;
201 * Mark a bio as coming in through the mpath node.
203 #define REQ_NVME_MPATH REQ_DRV
206 NVME_REQ_CANCELLED = (1 << 0),
207 NVME_REQ_USERCMD = (1 << 1),
208 NVME_MPATH_IO_STATS = (1 << 2),
209 NVME_MPATH_CNT_ACTIVE = (1 << 3),
212 static inline struct nvme_request *nvme_req(struct request *req)
214 return blk_mq_rq_to_pdu(req);
217 static inline u16 nvme_req_qid(struct request *req)
219 if (!req->q->queuedata)
222 return req->mq_hctx->queue_num + 1;
225 /* The below value is the specific amount of delay needed before checking
226 * readiness in case of the PCI_DEVICE(0x1c58, 0x0003), which needs the
227 * NVME_QUIRK_DELAY_BEFORE_CHK_RDY quirk enabled. The value (in ms) was
230 #define NVME_QUIRK_DELAY_AMOUNT 2300
233 * enum nvme_ctrl_state: Controller state
235 * @NVME_CTRL_NEW: New controller just allocated, initial state
236 * @NVME_CTRL_LIVE: Controller is connected and I/O capable
237 * @NVME_CTRL_RESETTING: Controller is resetting (or scheduled reset)
238 * @NVME_CTRL_CONNECTING: Controller is disconnected, now connecting the
240 * @NVME_CTRL_DELETING: Controller is deleting (or scheduled deletion)
241 * @NVME_CTRL_DELETING_NOIO: Controller is deleting and I/O is not
242 * disabled/failed immediately. This state comes
243 * after all async event processing took place and
244 * before ns removal and the controller deletion
246 * @NVME_CTRL_DEAD: Controller is non-present/unresponsive during
247 * shutdown or removal. In this case we forcibly
248 * kill all inflight I/O as they have no chance to
251 enum nvme_ctrl_state {
255 NVME_CTRL_CONNECTING,
257 NVME_CTRL_DELETING_NOIO,
261 struct nvme_fault_inject {
262 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
263 struct fault_attr attr;
264 struct dentry *parent;
265 bool dont_retry; /* DNR, do not retry */
266 u16 status; /* status code */
270 enum nvme_ctrl_flags {
271 NVME_CTRL_FAILFAST_EXPIRED = 0,
272 NVME_CTRL_ADMIN_Q_STOPPED = 1,
273 NVME_CTRL_STARTED_ONCE = 2,
274 NVME_CTRL_STOPPED = 3,
275 NVME_CTRL_SKIP_ID_CNS_CS = 4,
276 NVME_CTRL_DIRTY_CAPABILITY = 5,
277 NVME_CTRL_FROZEN = 6,
283 bool passthru_err_log_enabled;
284 enum nvme_ctrl_state state;
286 struct mutex scan_lock;
287 const struct nvme_ctrl_ops *ops;
288 struct request_queue *admin_q;
289 struct request_queue *connect_q;
290 struct request_queue *fabrics_q;
294 struct blk_mq_tag_set *tagset;
295 struct blk_mq_tag_set *admin_tagset;
296 struct list_head namespaces;
297 struct mutex namespaces_lock;
298 struct srcu_struct srcu;
299 struct device ctrl_device;
300 struct device *device; /* char device */
301 #ifdef CONFIG_NVME_HWMON
302 struct device *hwmon_device;
305 struct work_struct reset_work;
306 struct work_struct delete_work;
307 wait_queue_head_t state_wq;
309 struct nvme_subsystem *subsys;
310 struct list_head subsys_entry;
312 struct opal_dev *opal_dev;
323 u32 max_integrity_segments;
324 u32 max_zeroes_sectors;
325 #ifdef CONFIG_BLK_DEV_ZONED
335 atomic_t abort_limit;
347 unsigned int shutdown_timeout;
350 unsigned long quirks;
351 struct nvme_id_power_state psd[32];
352 struct nvme_effects_log *effects;
354 struct work_struct scan_work;
355 struct work_struct async_event_work;
356 struct delayed_work ka_work;
357 struct delayed_work failfast_work;
358 struct nvme_command ka_cmd;
359 unsigned long ka_last_check_time;
360 struct work_struct fw_act_work;
361 unsigned long events;
363 #ifdef CONFIG_NVME_MULTIPATH
364 /* asymmetric namespace access: */
369 struct mutex ana_lock;
370 struct nvme_ana_rsp_hdr *ana_log_buf;
372 struct timer_list anatt_timer;
373 struct work_struct ana_work;
377 #ifdef CONFIG_NVME_HOST_AUTH
378 struct work_struct dhchap_auth_work;
379 struct mutex dhchap_auth_mutex;
380 struct nvme_dhchap_queue_context *dhchap_ctxs;
381 struct nvme_dhchap_key *host_key;
382 struct nvme_dhchap_key *ctrl_key;
385 key_serial_t tls_pskid;
387 /* Power saving configuration */
388 u64 ps_max_latency_us;
404 struct nvmf_ctrl_options *opts;
406 struct page *discard_page;
407 unsigned long discard_page_busy;
409 struct nvme_fault_inject fault_inject;
411 enum nvme_ctrl_type cntrltype;
412 enum nvme_dctype dctype;
415 static inline enum nvme_ctrl_state nvme_ctrl_state(struct nvme_ctrl *ctrl)
417 return READ_ONCE(ctrl->state);
426 struct nvme_subsystem {
430 * Because we unregister the device on the last put we need
431 * a separate refcount.
434 struct list_head entry;
436 struct list_head ctrls;
437 struct list_head nsheads;
438 char subnqn[NVMF_NQN_SIZE];
441 char firmware_rev[8];
443 enum nvme_subsys_type subtype;
445 u16 awupf; /* 0's based value. */
447 #ifdef CONFIG_NVME_MULTIPATH
448 enum nvme_iopolicy iopolicy;
453 * Container structure for uniqueue namespace identifiers.
463 * Anchor structure for namespaces. There is one for each namespace in a
464 * NVMe subsystem that any of our controllers can see, and the namespace
465 * structure for each controller is chained of it. For private namespaces
466 * there is a 1:1 relation to our namespace structures, that is ->list
467 * only ever has a single entry for private namespaces.
469 struct nvme_ns_head {
470 struct list_head list;
471 struct srcu_struct srcu;
472 struct nvme_subsystem *subsys;
473 struct nvme_ns_ids ids;
479 struct list_head entry;
483 bool passthru_err_log_enabled;
484 struct nvme_effects_log *effects;
488 #ifdef CONFIG_BLK_DEV_ZONED
491 unsigned long features;
493 struct ratelimit_state rs_nuse;
496 struct device cdev_device;
498 struct gendisk *disk;
502 #ifdef CONFIG_NVME_MULTIPATH
503 struct bio_list requeue_list;
504 spinlock_t requeue_lock;
505 struct work_struct requeue_work;
506 struct work_struct partition_scan_work;
509 struct delayed_work remove_work;
510 unsigned int delayed_removal_secs;
511 #define NVME_NSHEAD_DISK_LIVE 0
512 #define NVME_NSHEAD_QUEUE_IF_NO_PATH 1
513 struct nvme_ns __rcu *current_path[];
517 static inline bool nvme_ns_head_multipath(struct nvme_ns_head *head)
519 return IS_ENABLED(CONFIG_NVME_MULTIPATH) && head->disk;
522 enum nvme_ns_features {
523 NVME_NS_EXT_LBAS = 1 << 0, /* support extended LBA format */
524 NVME_NS_METADATA_SUPPORTED = 1 << 1, /* support getting generated md */
525 NVME_NS_DEAC = 1 << 2, /* DEAC bit in Write Zeroes supported */
529 struct list_head list;
531 struct nvme_ctrl *ctrl;
532 struct request_queue *queue;
533 struct gendisk *disk;
534 #ifdef CONFIG_NVME_MULTIPATH
535 enum nvme_ana_state ana_state;
538 struct list_head siblings;
540 struct nvme_ns_head *head;
543 #define NVME_NS_REMOVING 0
544 #define NVME_NS_ANA_PENDING 2
545 #define NVME_NS_FORCE_RO 3
546 #define NVME_NS_READY 4
547 #define NVME_NS_SYSFS_ATTR_LINK 5
550 struct device cdev_device;
552 struct nvme_fault_inject fault_inject;
555 /* NVMe ns supports metadata actions by the controller (generate/strip) */
556 static inline bool nvme_ns_has_pi(struct nvme_ns_head *head)
558 return head->pi_type && head->ms == head->pi_size;
561 struct nvme_ctrl_ops {
563 struct module *module;
565 #define NVME_F_FABRICS (1 << 0)
566 #define NVME_F_METADATA_SUPPORTED (1 << 1)
567 #define NVME_F_BLOCKING (1 << 2)
569 const struct attribute_group **dev_attr_groups;
570 int (*reg_read32)(struct nvme_ctrl *ctrl, u32 off, u32 *val);
571 int (*reg_write32)(struct nvme_ctrl *ctrl, u32 off, u32 val);
572 int (*reg_read64)(struct nvme_ctrl *ctrl, u32 off, u64 *val);
573 void (*free_ctrl)(struct nvme_ctrl *ctrl);
574 void (*submit_async_event)(struct nvme_ctrl *ctrl);
575 int (*subsystem_reset)(struct nvme_ctrl *ctrl);
576 void (*delete_ctrl)(struct nvme_ctrl *ctrl);
577 void (*stop_ctrl)(struct nvme_ctrl *ctrl);
578 int (*get_address)(struct nvme_ctrl *ctrl, char *buf, int size);
579 void (*print_device_info)(struct nvme_ctrl *ctrl);
580 bool (*supports_pci_p2pdma)(struct nvme_ctrl *ctrl);
584 * nvme command_id is constructed as such:
585 * | xxxx | xxxxxxxxxxxx |
588 #define nvme_genctr_mask(gen) (gen & 0xf)
589 #define nvme_cid_install_genctr(gen) (nvme_genctr_mask(gen) << 12)
590 #define nvme_genctr_from_cid(cid) ((cid & 0xf000) >> 12)
591 #define nvme_tag_from_cid(cid) (cid & 0xfff)
593 static inline u16 nvme_cid(struct request *rq)
595 return nvme_cid_install_genctr(nvme_req(rq)->genctr) | rq->tag;
598 static inline struct request *nvme_find_rq(struct blk_mq_tags *tags,
601 u8 genctr = nvme_genctr_from_cid(command_id);
602 u16 tag = nvme_tag_from_cid(command_id);
605 rq = blk_mq_tag_to_rq(tags, tag);
607 pr_err("could not locate request for tag %#x\n",
611 if (unlikely(nvme_genctr_mask(nvme_req(rq)->genctr) != genctr)) {
612 dev_err(nvme_req(rq)->ctrl->device,
613 "request %#x genctr mismatch (got %#x expected %#x)\n",
614 tag, genctr, nvme_genctr_mask(nvme_req(rq)->genctr));
620 static inline struct request *nvme_cid_to_rq(struct blk_mq_tags *tags,
623 return blk_mq_tag_to_rq(tags, nvme_tag_from_cid(command_id));
627 * Return the length of the string without the space padding
629 static inline int nvme_strlen(char *s, int len)
631 while (s[len - 1] == ' ')
636 static inline void nvme_print_device_info(struct nvme_ctrl *ctrl)
638 struct nvme_subsystem *subsys = ctrl->subsys;
640 if (ctrl->ops->print_device_info) {
641 ctrl->ops->print_device_info(ctrl);
645 dev_err(ctrl->device,
646 "VID:%04x model:%.*s firmware:%.*s\n", subsys->vendor_id,
647 nvme_strlen(subsys->model, sizeof(subsys->model)),
648 subsys->model, nvme_strlen(subsys->firmware_rev,
649 sizeof(subsys->firmware_rev)),
650 subsys->firmware_rev);
653 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
654 void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
655 const char *dev_name);
656 void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inject);
657 void nvme_should_fail(struct request *req);
659 static inline void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
660 const char *dev_name)
663 static inline void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inj)
666 static inline void nvme_should_fail(struct request *req) {}
669 bool nvme_wait_reset(struct nvme_ctrl *ctrl);
670 int nvme_try_sched_reset(struct nvme_ctrl *ctrl);
672 static inline int nvme_reset_subsystem(struct nvme_ctrl *ctrl)
674 if (!ctrl->subsystem || !ctrl->ops->subsystem_reset)
676 return ctrl->ops->subsystem_reset(ctrl);
680 * Convert a 512B sector number to a device logical block number.
682 static inline u64 nvme_sect_to_lba(struct nvme_ns_head *head, sector_t sector)
684 return sector >> (head->lba_shift - SECTOR_SHIFT);
688 * Convert a device logical block number to a 512B sector number.
690 static inline sector_t nvme_lba_to_sect(struct nvme_ns_head *head, u64 lba)
692 return lba << (head->lba_shift - SECTOR_SHIFT);
696 * Convert byte length to nvme's 0-based num dwords
698 static inline u32 nvme_bytes_to_numd(size_t len)
700 return (len >> 2) - 1;
703 static inline bool nvme_is_ana_error(u16 status)
705 switch (status & NVME_SCT_SC_MASK) {
706 case NVME_SC_ANA_TRANSITION:
707 case NVME_SC_ANA_INACCESSIBLE:
708 case NVME_SC_ANA_PERSISTENT_LOSS:
715 static inline bool nvme_is_path_error(u16 status)
717 /* check for a status code type of 'path related status' */
718 return (status & NVME_SCT_MASK) == NVME_SCT_PATH;
722 * Fill in the status and result information from the CQE, and then figure out
723 * if blk-mq will need to use IPI magic to complete the request, and if yes do
724 * so. If not let the caller complete the request without an indirect function
727 static inline bool nvme_try_complete_req(struct request *req, __le16 status,
728 union nvme_result result)
730 struct nvme_request *rq = nvme_req(req);
731 struct nvme_ctrl *ctrl = rq->ctrl;
733 if (!(ctrl->quirks & NVME_QUIRK_SKIP_CID_GEN))
736 rq->status = le16_to_cpu(status) >> 1;
738 /* inject error when permitted by fault injection framework */
739 nvme_should_fail(req);
740 if (unlikely(blk_should_fake_timeout(req->q)))
742 return blk_mq_complete_request_remote(req);
745 static inline void nvme_get_ctrl(struct nvme_ctrl *ctrl)
747 get_device(ctrl->device);
750 static inline void nvme_put_ctrl(struct nvme_ctrl *ctrl)
752 put_device(ctrl->device);
755 static inline bool nvme_is_aen_req(u16 qid, __u16 command_id)
758 nvme_tag_from_cid(command_id) >= NVME_AQ_BLK_MQ_DEPTH;
762 * Returns true for sink states that can't ever transition back to live.
764 static inline bool nvme_state_terminal(struct nvme_ctrl *ctrl)
766 switch (nvme_ctrl_state(ctrl)) {
769 case NVME_CTRL_RESETTING:
770 case NVME_CTRL_CONNECTING:
772 case NVME_CTRL_DELETING:
773 case NVME_CTRL_DELETING_NOIO:
777 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
782 void nvme_end_req(struct request *req);
783 void nvme_complete_rq(struct request *req);
784 void nvme_complete_batch_req(struct request *req);
786 static __always_inline void nvme_complete_batch(struct io_comp_batch *iob,
787 void (*fn)(struct request *rq))
791 rq_list_for_each(&iob->req_list, req) {
793 nvme_complete_batch_req(req);
795 blk_mq_end_request_batch(iob);
798 blk_status_t nvme_host_path_error(struct request *req);
799 bool nvme_cancel_request(struct request *req, void *data);
800 void nvme_cancel_tagset(struct nvme_ctrl *ctrl);
801 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl);
802 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
803 enum nvme_ctrl_state new_state);
804 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, bool shutdown);
805 int nvme_enable_ctrl(struct nvme_ctrl *ctrl);
806 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
807 const struct nvme_ctrl_ops *ops, unsigned long quirks);
808 int nvme_add_ctrl(struct nvme_ctrl *ctrl);
809 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl);
810 void nvme_start_ctrl(struct nvme_ctrl *ctrl);
811 void nvme_stop_ctrl(struct nvme_ctrl *ctrl);
812 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl, bool was_suspended);
813 int nvme_alloc_admin_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
814 const struct blk_mq_ops *ops, unsigned int cmd_size);
815 void nvme_remove_admin_tag_set(struct nvme_ctrl *ctrl);
816 int nvme_alloc_io_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
817 const struct blk_mq_ops *ops, unsigned int nr_maps,
818 unsigned int cmd_size);
819 void nvme_remove_io_tag_set(struct nvme_ctrl *ctrl);
821 void nvme_remove_namespaces(struct nvme_ctrl *ctrl);
823 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
824 volatile union nvme_result *res);
826 void nvme_quiesce_io_queues(struct nvme_ctrl *ctrl);
827 void nvme_unquiesce_io_queues(struct nvme_ctrl *ctrl);
828 void nvme_quiesce_admin_queue(struct nvme_ctrl *ctrl);
829 void nvme_unquiesce_admin_queue(struct nvme_ctrl *ctrl);
830 void nvme_mark_namespaces_dead(struct nvme_ctrl *ctrl);
831 void nvme_sync_queues(struct nvme_ctrl *ctrl);
832 void nvme_sync_io_queues(struct nvme_ctrl *ctrl);
833 void nvme_unfreeze(struct nvme_ctrl *ctrl);
834 void nvme_wait_freeze(struct nvme_ctrl *ctrl);
835 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout);
836 void nvme_start_freeze(struct nvme_ctrl *ctrl);
838 static inline enum req_op nvme_req_op(struct nvme_command *cmd)
840 return nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
843 #define NVME_QID_ANY -1
844 void nvme_init_request(struct request *req, struct nvme_command *cmd);
845 void nvme_cleanup_cmd(struct request *req);
846 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req);
847 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
848 struct request *req);
849 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
850 bool queue_live, enum nvme_ctrl_state state);
852 static inline bool nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
855 enum nvme_ctrl_state state = nvme_ctrl_state(ctrl);
857 if (likely(state == NVME_CTRL_LIVE))
859 if (ctrl->ops->flags & NVME_F_FABRICS && state == NVME_CTRL_DELETING)
861 return __nvme_check_ready(ctrl, rq, queue_live, state);
865 * NSID shall be unique for all shared namespaces, or if at least one of the
866 * following conditions is met:
867 * 1. Namespace Management is supported by the controller
868 * 2. ANA is supported by the controller
869 * 3. NVM Set are supported by the controller
871 * In other case, private namespace are not required to report a unique NSID.
873 static inline bool nvme_is_unique_nsid(struct nvme_ctrl *ctrl,
874 struct nvme_ns_head *head)
876 return head->shared ||
877 (ctrl->oacs & NVME_CTRL_OACS_NS_MNGT_SUPP) ||
878 (ctrl->subsys->cmic & NVME_CTRL_CMIC_ANA) ||
879 (ctrl->ctratt & NVME_CTRL_CTRATT_NVM_SETS);
883 * Flags for __nvme_submit_sync_cmd()
885 typedef __u32 __bitwise nvme_submit_flags_t;
888 /* Insert request at the head of the queue */
889 NVME_SUBMIT_AT_HEAD = (__force nvme_submit_flags_t)(1 << 0),
890 /* Set BLK_MQ_REQ_NOWAIT when allocating request */
891 NVME_SUBMIT_NOWAIT = (__force nvme_submit_flags_t)(1 << 1),
892 /* Set BLK_MQ_REQ_RESERVED when allocating request */
893 NVME_SUBMIT_RESERVED = (__force nvme_submit_flags_t)(1 << 2),
894 /* Retry command when NVME_STATUS_DNR is not set in the result */
895 NVME_SUBMIT_RETRY = (__force nvme_submit_flags_t)(1 << 3),
898 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
899 void *buf, unsigned bufflen);
900 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
901 union nvme_result *result, void *buffer, unsigned bufflen,
902 int qid, nvme_submit_flags_t flags);
903 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
904 unsigned int dword11, void *buffer, size_t buflen,
906 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
907 unsigned int dword11, void *buffer, size_t buflen,
909 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count);
910 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl);
911 int nvme_reset_ctrl(struct nvme_ctrl *ctrl);
912 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl);
913 int nvme_delete_ctrl(struct nvme_ctrl *ctrl);
914 void nvme_queue_scan(struct nvme_ctrl *ctrl);
915 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
916 void *log, size_t size, u64 offset);
917 bool nvme_tryget_ns_head(struct nvme_ns_head *head);
918 void nvme_put_ns_head(struct nvme_ns_head *head);
919 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
920 const struct file_operations *fops, struct module *owner);
921 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device);
922 int nvme_ioctl(struct block_device *bdev, blk_mode_t mode,
923 unsigned int cmd, unsigned long arg);
924 long nvme_ns_chr_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
925 int nvme_ns_head_ioctl(struct block_device *bdev, blk_mode_t mode,
926 unsigned int cmd, unsigned long arg);
927 long nvme_ns_head_chr_ioctl(struct file *file, unsigned int cmd,
929 long nvme_dev_ioctl(struct file *file, unsigned int cmd,
931 int nvme_ns_chr_uring_cmd_iopoll(struct io_uring_cmd *ioucmd,
932 struct io_comp_batch *iob, unsigned int poll_flags);
933 int nvme_ns_chr_uring_cmd(struct io_uring_cmd *ioucmd,
934 unsigned int issue_flags);
935 int nvme_ns_head_chr_uring_cmd(struct io_uring_cmd *ioucmd,
936 unsigned int issue_flags);
937 int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
938 struct nvme_id_ns **id);
939 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo);
940 int nvme_dev_uring_cmd(struct io_uring_cmd *ioucmd, unsigned int issue_flags);
942 extern const struct attribute_group *nvme_ns_attr_groups[];
943 extern const struct attribute_group nvme_ns_mpath_attr_group;
944 extern const struct pr_ops nvme_pr_ops;
945 extern const struct block_device_operations nvme_ns_head_ops;
946 extern const struct attribute_group nvme_dev_attrs_group;
947 extern const struct attribute_group *nvme_subsys_attrs_groups[];
948 extern const struct attribute_group *nvme_dev_attr_groups[];
949 extern const struct block_device_operations nvme_bdev_ops;
951 void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl);
952 struct nvme_ns *nvme_find_path(struct nvme_ns_head *head);
953 #ifdef CONFIG_NVME_MULTIPATH
954 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
956 return ctrl->ana_log_buf != NULL;
959 void nvme_mpath_unfreeze(struct nvme_subsystem *subsys);
960 void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys);
961 void nvme_mpath_start_freeze(struct nvme_subsystem *subsys);
962 void nvme_mpath_default_iopolicy(struct nvme_subsystem *subsys);
963 void nvme_failover_req(struct request *req);
964 void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl);
965 int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,struct nvme_ns_head *head);
966 void nvme_mpath_add_sysfs_link(struct nvme_ns_head *ns);
967 void nvme_mpath_remove_sysfs_link(struct nvme_ns *ns);
968 void nvme_mpath_add_disk(struct nvme_ns *ns, __le32 anagrpid);
969 void nvme_mpath_put_disk(struct nvme_ns_head *head);
970 int nvme_mpath_init_identify(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id);
971 void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl);
972 void nvme_mpath_update(struct nvme_ctrl *ctrl);
973 void nvme_mpath_uninit(struct nvme_ctrl *ctrl);
974 void nvme_mpath_stop(struct nvme_ctrl *ctrl);
975 bool nvme_mpath_clear_current_path(struct nvme_ns *ns);
976 void nvme_mpath_revalidate_paths(struct nvme_ns *ns);
977 void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl);
978 void nvme_mpath_remove_disk(struct nvme_ns_head *head);
979 void nvme_mpath_start_request(struct request *rq);
980 void nvme_mpath_end_request(struct request *rq);
982 static inline void nvme_trace_bio_complete(struct request *req)
984 struct nvme_ns *ns = req->q->queuedata;
986 if ((req->cmd_flags & REQ_NVME_MPATH) && req->bio)
987 trace_block_bio_complete(ns->head->disk->queue, req->bio);
990 extern bool multipath;
991 extern struct device_attribute dev_attr_ana_grpid;
992 extern struct device_attribute dev_attr_ana_state;
993 extern struct device_attribute dev_attr_queue_depth;
994 extern struct device_attribute dev_attr_numa_nodes;
995 extern struct device_attribute dev_attr_delayed_removal_secs;
996 extern struct device_attribute subsys_attr_iopolicy;
998 static inline bool nvme_disk_is_ns_head(struct gendisk *disk)
1000 return disk->fops == &nvme_ns_head_ops;
1002 static inline bool nvme_mpath_queue_if_no_path(struct nvme_ns_head *head)
1004 if (test_bit(NVME_NSHEAD_QUEUE_IF_NO_PATH, &head->flags))
1009 #define multipath false
1010 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
1014 static inline void nvme_failover_req(struct request *req)
1017 static inline void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl)
1020 static inline int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,
1021 struct nvme_ns_head *head)
1025 static inline void nvme_mpath_add_disk(struct nvme_ns *ns, __le32 anagrpid)
1028 static inline void nvme_mpath_put_disk(struct nvme_ns_head *head)
1031 static inline void nvme_mpath_add_sysfs_link(struct nvme_ns *ns)
1034 static inline void nvme_mpath_remove_sysfs_link(struct nvme_ns *ns)
1037 static inline bool nvme_mpath_clear_current_path(struct nvme_ns *ns)
1041 static inline void nvme_mpath_revalidate_paths(struct nvme_ns *ns)
1044 static inline void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl)
1047 static inline void nvme_mpath_remove_disk(struct nvme_ns_head *head)
1050 static inline void nvme_trace_bio_complete(struct request *req)
1053 static inline void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl)
1056 static inline int nvme_mpath_init_identify(struct nvme_ctrl *ctrl,
1057 struct nvme_id_ctrl *id)
1059 if (ctrl->subsys->cmic & NVME_CTRL_CMIC_ANA)
1060 dev_warn(ctrl->device,
1061 "Please enable CONFIG_NVME_MULTIPATH for full support of multi-port devices.\n");
1064 static inline void nvme_mpath_update(struct nvme_ctrl *ctrl)
1067 static inline void nvme_mpath_uninit(struct nvme_ctrl *ctrl)
1070 static inline void nvme_mpath_stop(struct nvme_ctrl *ctrl)
1073 static inline void nvme_mpath_unfreeze(struct nvme_subsystem *subsys)
1076 static inline void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys)
1079 static inline void nvme_mpath_start_freeze(struct nvme_subsystem *subsys)
1082 static inline void nvme_mpath_default_iopolicy(struct nvme_subsystem *subsys)
1085 static inline void nvme_mpath_start_request(struct request *rq)
1088 static inline void nvme_mpath_end_request(struct request *rq)
1091 static inline bool nvme_disk_is_ns_head(struct gendisk *disk)
1095 static inline bool nvme_mpath_queue_if_no_path(struct nvme_ns_head *head)
1099 #endif /* CONFIG_NVME_MULTIPATH */
1101 int nvme_ns_get_unique_id(struct nvme_ns *ns, u8 id[16],
1102 enum blk_unique_id type);
1104 struct nvme_zone_info {
1106 unsigned int max_open_zones;
1107 unsigned int max_active_zones;
1110 int nvme_ns_report_zones(struct nvme_ns *ns, sector_t sector,
1111 unsigned int nr_zones, report_zones_cb cb, void *data);
1112 int nvme_query_zone_info(struct nvme_ns *ns, unsigned lbaf,
1113 struct nvme_zone_info *zi);
1114 void nvme_update_zone_info(struct nvme_ns *ns, struct queue_limits *lim,
1115 struct nvme_zone_info *zi);
1116 #ifdef CONFIG_BLK_DEV_ZONED
1117 blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns, struct request *req,
1118 struct nvme_command *cmnd,
1119 enum nvme_zone_mgmt_action action);
1121 static inline blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns,
1122 struct request *req, struct nvme_command *cmnd,
1123 enum nvme_zone_mgmt_action action)
1125 return BLK_STS_NOTSUPP;
1129 static inline struct nvme_ns *nvme_get_ns_from_dev(struct device *dev)
1131 struct gendisk *disk = dev_to_disk(dev);
1133 WARN_ON(nvme_disk_is_ns_head(disk));
1134 return disk->private_data;
1137 #ifdef CONFIG_NVME_HWMON
1138 int nvme_hwmon_init(struct nvme_ctrl *ctrl);
1139 void nvme_hwmon_exit(struct nvme_ctrl *ctrl);
1141 static inline int nvme_hwmon_init(struct nvme_ctrl *ctrl)
1146 static inline void nvme_hwmon_exit(struct nvme_ctrl *ctrl)
1151 static inline void nvme_start_request(struct request *rq)
1153 if (rq->cmd_flags & REQ_NVME_MPATH)
1154 nvme_mpath_start_request(rq);
1155 blk_mq_start_request(rq);
1158 static inline bool nvme_ctrl_sgl_supported(struct nvme_ctrl *ctrl)
1160 return ctrl->sgls & (NVME_CTRL_SGLS_BYTE_ALIGNED |
1161 NVME_CTRL_SGLS_DWORD_ALIGNED);
1164 static inline bool nvme_ctrl_meta_sgl_supported(struct nvme_ctrl *ctrl)
1166 if (ctrl->ops->flags & NVME_F_FABRICS)
1168 return ctrl->sgls & NVME_CTRL_SGLS_MSDS;
1171 #ifdef CONFIG_NVME_HOST_AUTH
1172 int __init nvme_init_auth(void);
1173 void __exit nvme_exit_auth(void);
1174 int nvme_auth_init_ctrl(struct nvme_ctrl *ctrl);
1175 void nvme_auth_stop(struct nvme_ctrl *ctrl);
1176 int nvme_auth_negotiate(struct nvme_ctrl *ctrl, int qid);
1177 int nvme_auth_wait(struct nvme_ctrl *ctrl, int qid);
1178 void nvme_auth_free(struct nvme_ctrl *ctrl);
1179 void nvme_auth_revoke_tls_key(struct nvme_ctrl *ctrl);
1181 static inline int nvme_auth_init_ctrl(struct nvme_ctrl *ctrl)
1185 static inline int __init nvme_init_auth(void)
1189 static inline void __exit nvme_exit_auth(void)
1192 static inline void nvme_auth_stop(struct nvme_ctrl *ctrl) {};
1193 static inline int nvme_auth_negotiate(struct nvme_ctrl *ctrl, int qid)
1195 return -EPROTONOSUPPORT;
1197 static inline int nvme_auth_wait(struct nvme_ctrl *ctrl, int qid)
1199 return -EPROTONOSUPPORT;
1201 static inline void nvme_auth_free(struct nvme_ctrl *ctrl) {};
1202 static inline void nvme_auth_revoke_tls_key(struct nvme_ctrl *ctrl) {};
1205 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1207 u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode);
1208 int nvme_execute_rq(struct request *rq, bool at_head);
1209 void nvme_passthru_end(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u32 effects,
1210 struct nvme_command *cmd, int status);
1211 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file);
1212 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid);
1213 bool nvme_get_ns(struct nvme_ns *ns);
1214 void nvme_put_ns(struct nvme_ns *ns);
1216 static inline bool nvme_multi_css(struct nvme_ctrl *ctrl)
1218 return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI;
1221 #endif /* _NVME_H */